Certainly, if implementations never matured to market, a winter *would* ensue. :) Yet! What I pointed-out regarding “error rates” is key to a *viable* product. For example, if an Amazon picker-robot has an error-rate of 1%, then a human will need to intervene every few minutes! That isn’t worth the expense and re-organization costs. So, even though improvements in performance seem to have ‘slowed down’, it’s actually the case that moving from ‘99% accuracy’ to ‘99.5%’ cuts errors in half, which is the difference between commercialization or winter. Companies are focused on that last, insignificant-seeming margin, because that’s when you finally reach the prize!